Auditing Claim Data for Aggregate Risk

Why Claim-Data Audits Come First

An aggregate-risk calculation can be arithmetically correct and still describe the wrong risk. The problem may begin with a claim counted twice, exposure measured over a different period from the claims, or loss amounts that mix gross losses with insurer payments. Before selecting a frequency or severity model, an actuary must establish what each record represents and whether its values share a consistent basis.

This tutorial develops a practical audit framework for hypothetical claim data. It focuses on definitions, reconciliation, and coverage interpretation, not on fitting a frequency or severity distribution. Unless a problem explicitly specifies otherwise, the examples use dollars, make no mortality assumption, and assume no parametric probability distribution. The audit principles are aligned with the SOA Exam FAM (Fundamentals of Actuarial Mathematics) syllabus.

Definition: An exposure unit is the unit over which an insured risk is observed, such as one policy-year or one vehicle-year. The observation period is the time window to which the data are intended to relate. A claim count is a count of eligible claims under a stated counting rule, not automatically a count of database rows. A loss amount must be identified by its basis, such as gross loss, paid amount, case-incurred amount, or insurer payment after coverage modifications.

These units are related but are not interchangeable. One policy may contribute less than a full policy-year if it is active for only part of the observation period. One claim may generate multiple payment transactions, and one claim record may show no payment yet. A claim might also be reported or paid in a different period from the one in which the loss occurred. The audit must state which dates determine inclusion.

Specify the Data Before Adding It

A useful audit begins by writing down the intended basis for each quantity. For example, claims may be counted by date of occurrence, while payments are summarized by transaction date. Those choices can be valid for different purposes, but they do not automatically produce matching totals. State the basis before comparing numbers.

Exposure: What is the unit, and how much of it was observed? Examples include policy-years or vehicle-years.
Period: Are claims included by occurrence date, report date, or another specified date? What is the valuation date for payments and reserves?
Claim count: Does one count represent a claim identifier, an occurrence, a claimant, or a payment transaction?
Loss amount: Is the recorded amount gross loss, amount paid to date, case reserve, or total insurer payment after coverage terms?
Coverage: What deductible, coinsurance, limit, or other modification applies, and in what order?

An audit also checks keys and status fields. Duplicate claim identifiers may indicate repeated transaction rows rather than duplicate claims. Conversely, two distinct records can describe the same occurrence if the data use inconsistent identifiers. A zero-paid or open claim is not necessarily a zero-loss claim. These are data questions to resolve before modeling, rather than assumptions to hide in a calculation.

Worked Example: Reconciling Claim IDs and Payment Transactions

A hypothetical insurer observes four policies during calendar year 2025. Policy A is active for all 12 months, policy B for 9 months, policy C for 6 months, and policy D for 3 months. The intended exposure unit is one policy-year. Claim data are valued at December 31, 2025, and claims are included only when their occurrence dates fall in 2025. The ledger contains these records: claim A1 has two payment transactions of $600 and $400; claim A2 has paid $0 and a case reserve of $500; claim B1 has a payment transaction of $700; claim D1 has a payment of $300 but an occurrence date of January 3, 2026. Reconcile exposure, included claim count, paid amount, and case-incurred amount. Assume the stated records are complete for 2025 claims, with no additional reserve component to include.

Method. First total exposure in policy-years, then determine eligible unique claim identifiers by occurrence date, and finally sum the appropriate financial fields. No frequency or severity distribution is assumed.

Exposure. Convert each policy’s observed months to a fraction of a year:

1+\frac{9}{12}+\frac{6}{12}+\frac{3}{12} =1+0.75+0.50+0.25 =2.50\text{ policy-years}.

Claim count. A1, A2, and B1 have 2025 occurrence dates, so there are 3 eligible unique claims. D1 occurred in 2026 and is excluded under the stated rule. The five ledger rows are not five claims: A1 appears twice because it has two transactions.

Paid amount. Add all 2025 payment transactions, including the two distinct A1 transactions:

600+400+0+700=\$1{,}700.

Case-incurred amount. For this example, case-incurred amount at the valuation date is paid to date plus the stated case reserve. Therefore:

\$1{,}700+\$500=\$2{,}200.

Check. The reconciled figures are 2.50 policy-years, 3 eligible claims, $1,700 paid, and $2,200 case-incurred. D1 contributes to none of these 2025 claim totals under the occurrence-date rule. This avoids the common mistake of treating transaction rows as claims or including a next-year occurrence just because it appears in the ledger.

Reconcile Gross Losses and Covered Payments

The amount attached to a claim is meaningful only after its coverage basis is identified. A gross loss describes the loss before the specified insurance payment calculation. An insurer payment reflects the contract terms and may be smaller because of a deductible, coinsurance, or limit. The same claim can therefore have several valid amounts, but they cannot be added or compared as if they were the same variable.

Definition: For a gross loss $x\geq 0$, a per-claim deductible $d\geq 0$, and a maximum insurer payment $u\geq 0$, the payment under a deductible followed by a payment cap is $$ Y^L=\min\bigl((x-d)_+,u\bigr),\qquad (z)_+=\max(z,0). $$ The formula applies only when the stated contract has this order and the cap is a maximum payment per claim. It does not describe an aggregate limit shared across claims or a different ordering of coverage modifications.

A concise justification follows directly from the contract. The deductible leaves no payment when $x\leq d$, and otherwise leaves $x-d$ before the cap. The insurer pays that nonnegative amount unless it exceeds $u$, in which case payment is $u$. Thus the payment is the smaller of the post-deductible amount and the per-claim maximum.

Worked Example: Auditing Payments Under a Deductible and Per-Claim Cap

A hypothetical policy covers three claims with gross losses of $250, $1,300, and $2,100. Each claim has a $300 deductible, and the insurer’s payment is capped at $1,500 per claim. All amounts are in dollars; there are no other coverage modifications. Reconcile the claim count, gross total, and insurer-payment total.

Method. Apply the coverage formula separately to each unique claim. The deductible and cap are per claim, not applied once to the portfolio total.

For the $250 loss:

\min\bigl((250-300)_+,1500\bigr) =\min(0,1500)=\$0.

For the $1,300 loss:

\min\bigl((1300-300)_+,1500\bigr) =\min(1000,1500)=\$1{,}000.

For the $2,100 loss:

\min\bigl((2100-300)_+,1500\bigr) =\min(1800,1500)=\$1{,}500.

There are 3 claims. The gross-loss total is $250+1300+2100=\$3{,}650$, while the insurer-payment total is $0+1000+1500=\$2{,}500$.

Check and trap. The gross total less one deductible is $3,350, but this is not the insurer-payment total: the deductible applies to each claim and the cap also applies to each claim. Applying a single $1,500 cap to the portfolio would also be wrong. The claim-by-claim payments sum to $2,500, which is less than the gross total as required by these coverage terms.

If a data file contains only payments, a capped payment may not reveal the exact gross loss. Under a $300 deductible and $1,500 payment cap, for example, a $1,500 payment establishes that the gross loss was at least $1,800, but it does not determine the amount above that threshold. The correct audit conclusion is that the gross amount is not recoverable from that payment alone; do not silently treat the payment as an uncapped loss.

Worked Example: Reconciling Different Coverage Bases

A hypothetical portfolio has four claims, all occurring during the same observation period. Policy A has a $100 per-claim deductible and a $900 maximum payment per claim. Its gross losses are $800 and $1,500. Policy B has a $200 deductible, pays 80% of the amount remaining after the deductible, and has a $1,000 maximum payment per claim. Its gross losses are $600 and $1,600. Assume coinsurance is applied after the deductible and the payment cap is applied last. Reconcile gross losses and insurer payments without mixing the two bases.

Method. Count the four claims, calculate each payment under its own policy terms, and then sum gross amounts and payments separately. No probability distribution is assumed.

For Policy A, the $800 loss produces:

\min\bigl((800-100)_+,900\bigr)=\min(700,900)=\$700.

The $1,500 loss produces:

\min\bigl((1500-100)_+,900\bigr)=\min(1400,900)=\$900.

For Policy B, the $600 loss leaves $600-200=\$400$ after the deductible. Coinsurance gives $0.80(400)=\$320$, below the cap. The $1,600 loss leaves $1600-200=\$1,400$; coinsurance gives $0.80(1400)=\$1,120$, which is reduced to the $1,000 cap.

The gross-loss total is:

800+1500+600+1600=\$4{,}500.

The insurer-payment total is:

700+900+320+1000=\$2{,}920.

Check and trap. There are four claims, with $4,500 of gross loss and $2,920 of covered payments. Multiplying every post-deductible amount by 80% would be wrong because the 80% coinsurance applies only to Policy B. Likewise, adding the gross total to the covered-payment total would double-count different measures of the same claims.

Why the Aggregation Basis Matters

For a set of claims $1,\ldots,n$, let $x_j$ be the gross loss for claim $j$ and $y_j$ its insurer payment under the specified coverage. The portfolio payment is the sum of the claim payments. This follows because each claim contributes its own payment under a per-claim contract:

Y_{\mathrm{total}}=\sum_{j=1}^{n} y_j.

The expression is not generally equal to applying the coverage formula once to the total gross loss. A deductible per claim is charged separately, and a per-claim cap is applied separately. An aggregate policy limit is different: it can constrain the sum across claims. The audit must therefore label each term as per-claim or aggregate before producing a total.

Timing is just as important as coverage. A paid amount at a valuation date is the sum of eligible payment transactions posted by then. A case-incurred amount in a stated convention may be paid to date plus case reserves. If a task specifies an incurred basis that includes additional components, those components must also be reconciled; they should not be assumed to be present in a file that only records payments and case reserves.

A Practical Reconciliation Sequence

Set the target basis.
Write down the exposure unit, observation period, claim-inclusion date, valuation date, and whether amounts are gross losses or covered payments.

Check exposure records.
Confirm that each exposure unit is represented once for the appropriate portion of the period. Convert partial periods to the stated unit only when the data support that conversion.

Reconcile claim identities.
Use the stated claim-counting rule to distinguish unique eligible claims from transaction rows, duplicate records, and claims outside the period.

Reconcile amounts.
Sum payment transactions for paid amounts, add the stated reserves only for an incurred measure, and apply coverage terms to gross losses claim by claim when required.

Report unresolved limits.
Identify missing or non-recoverable information, such as gross losses hidden by payment caps. Do not fill gaps with unstated assumptions.

Common Audit Traps

Counting rows instead of claims. Payment transactions can create several rows for one claim. Count according to the chosen claim identifier and eligibility rule.
Using mismatched dates. A claim occurring within the period can be paid later; a claim paid within the period can have occurred earlier. State which date governs inclusion.
Treating zero paid as no claim. A reported claim may be open, denied, or not yet paid. A zero payment does not itself establish that the claim should be removed from the count.
Mixing gross and covered amounts. Gross losses and insurer payments answer different questions. Keep separate totals and labels.
Applying a per-claim term to a portfolio total. Deductibles and limits must be applied at the level specified by the contract. Do not replace a per-claim calculation with one calculation on aggregate loss.
Inferring unavailable amounts. When coverage caps truncate payments, a payment-only record may not identify the original loss exactly. Flag the limitation rather than inventing a value.

Key takeaway: Before using claim data for aggregate-risk work, reconcile the exposure denominator, the claim-counting rule, the observation and valuation dates, and the financial basis of every amount. A valid calculation starts with records that refer to the same risk, period, and coverage definition.

Check Your Understanding

For each problem, apply the stated observation and coverage rules before reconciling the totals.

A hypothetical ledger has two transactions of $250 and $150 for claim Q1, one $400 transaction for claim Q2, and one $90 transaction for claim Q3. All three claims occurred in the observation period. What are the unique claim count and total paid amount? (Answer: 3 claims and $890 paid)
Three policies contribute 12, 6, and 3 months of exposure. The claim file has four unique claim identifiers, but one occurred outside the specified observation period. How much exposure is present, and how many claims are eligible? (Answer: 1.75 policy-years and 3 eligible claims)
A gross loss is $1,250, with a $250 per-claim deductible and a $700 maximum insurer payment. What is the payment? (Answer: $700)
A claim has $480 paid to date and a $220 case reserve at the valuation date. Under the convention that case-incurred amount equals paid plus case reserve, what are the paid and case-incurred amounts? (Answer: $480 paid and $700 case-incurred)
Two claims have gross losses of $500 and $1,400. Each has a $100 deductible and 50% coinsurance after the deductible, with no limit. Reconcile the gross total and insurer-payment total. (Answer: $1,900 gross and $850 insurer payment)

What You'll Learn